任务【模型评估】
记录五个模型关于precision，rescore，f1，auc，roc的评分表格，画出auc和roc曲线图
遇到难题
实验代码
#!/usr/bin/env python 3.6
#-*- coding:utf-8 -*-
# @File : Model_evaluation.py
# @Date : 2018-11-20
# @Author : 黑桃
# @Software: PyCharm
import pickle
from matplotlib import pyplot as plt
from sklearn.externals import joblib
from sklearn.metrics import accuracy_score, recall_score, f1_score, roc_auc_score, roc_curve
path = "E:/MyPython/Machine_learning_GoGoGo/"
"""=====================================================================================================================
1 读取特征
"""
print("0 读取特征")
f = open(path + 'feature/feature_V1.pkl', 'rb')
train, test, y_train,y_test= pickle.load(f)
f.close()
"""=====================================================================================================================
2 读取模型
"""
print("1 读取模型")
SVM_linear = joblib.load( path + "model/SVM_linear.pkl")
SVM_poly = joblib.load( path + "model/SVM_poly.pkl")
SVM_rbf = joblib.load( path + "model/SVM_rbf.pkl")
SVM_sigmoid = joblib.load( path + "model/SVM_sigmoid.pkl")
lg_120 = joblib.load( path + "model/lg_120.pkl")
DT = joblib.load( path + "model/DT.pkl")
xgb_sklearn = joblib.load( path + "model/xgb_sklearn.pkl")
lgb_sklearn = joblib.load( path + "model/lgb_sklearn.pkl")
xgb = joblib.load( path + "model/xgb.pkl")
lgb = joblib.load( path + "model/lgb.pkl")
"""=====================================================================================================================
3 模型评估
"""
def model_evalua(clf, X_train, X_test, y_train, y_test,clf_name):
y_train_pred = clf.predict(X_train)
y_test_pred = clf.predict(X_test)
y_train_pred_proba = clf.predict_proba(X_train)[:, 1]
y_test_pred_proba = clf.predict_proba(X_test)[:, 1]
"""【AUC Score】"""
print('AUC Score')
print("Train_AUC Score ：{:.4f}".format(roc_auc_score(y_train, y_train_pred)))
print("Test_AUC Score ：{:.4f}".format(roc_auc_score(y_test, y_test_pred)))
"""【准确性】"""
print('准确性：')
print('Train_准确性：{:.4f}'.format(accuracy_score(y_train, y_train_pred)))
print('Test_准确性：{:.4f}'.format(accuracy_score(y_test, y_test_pred)))
"""【召回率】"""
print('召回率：')
print('Train_召回率：{:.4f}'.format(recall_score(y_train, y_train_pred)))
print('Test_召回率：{:.4f}'.format(recall_score(y_test, y_test_pred)))
"""【f1_score】"""
print('f1_score：')
print('Train_f1_score：{:.4f}'.format(f1_score(y_train, y_train_pred)))
print('Test_f1_score：{:.4f}'.format(f1_score(y_test, y_test_pred)))
#描绘 ROC 曲线
fpr_tr, tpr_tr, _ = roc_curve(y_train, y_train_pred_proba)
fpr_te, tpr_te, _ = roc_curve(y_test, y_test_pred_proba)
# KS
print('KS：')
print('Train：{:.4f}'.format(max(abs((fpr_tr - tpr_tr)))))
print('Test：{:.4f}'.format(max(abs((fpr_te - tpr_te)))))
plt.plot(fpr_tr, tpr_tr, 'r-',
label = "Train:AUC: {:.3f} KS:{:.3f}".format(roc_auc_score(y_train, y_train_pred_proba),
max(abs((fpr_tr - tpr_tr)))))
plt.plot(fpr_te, tpr_te, 'g-',
label="Test:AUC: {:.3f} KS:{:.3f}".format(roc_auc_score(y_test, y_test_pred_proba),
max(abs((fpr_tr - tpr_tr)))))
plt.plot([0, 1], [0, 1], 'd--')
plt.legend(loc='best')
plt.title(clf_name + "ROC curse")
plt.savefig(path +'picture/'+clf_name+'.jpg')
plt.show()
print('-------------------SVM_linear-------------------')
model_evalua(SVM_linear, train, test, y_train, y_test,'SVM_linear')
print('-------------------SVM_poly-------------------：')
model_evalua(SVM_poly, train, test, y_train, y_test,'SVM_poly')
print('-------------------SVM_rbf-------------------：')
model_evalua(SVM_rbf, train, test, y_train, y_test,'SVM_rbf')
print('-------------------SVM_sigmoid-------------------：')
model_evalua(SVM_sigmoid, train, test, y_train, y_test,'SVM_sigmoid')
print('-------------------lg_120-------------------')
model_evalua(lg_120, train, test, y_train, y_test,'lg_120')
print('-------------------DT-------------------')
model_evalua(DT, train, test, y_train, y_test,'DT')
print('-------------------xgb_sklearn-------------------')
model_evalua(xgb_sklearn, train, test, y_train, y_test,'xgb_sklearn')
# print('-------------------xgb-------------------')
# model_evalua(xgb, train, test, y_train, y_test)
print('-------------------lgb_sklearn-------------------')
model_evalua(lgb_sklearn, train, test, y_train, y_test,'lgb_sklearn')
# print('-------------------lgb-------------------')
# model_evalua(lgb, train, test, y_train, y_test)
实验结果
| precision | recall | f1_score | KS | ROC_AUC | ROC曲线 | ||
|---|---|---|---|---|---|---|---|
| SVM_linear | Train_准确性：0.7878Test_准确性：0.7442 | Train_召回率：0.1683Test_召回率：0.3377 | Train_f1_score：0.2781 Test_f1_score：0.4160 | Train：0.4519 Test：0.2590 | Train_AUC Score ：0.5774 Test_AUC Score ：0.6160 |  |
|
| SVM_poly | Train_准确性：0.7815 Test_准确性：0.7267 | Train_召回率：0.1027 Test_召回率：0.0597 | Train_f1_score：0.1859 Test_f1_score：0.1055 | Train：0.7099 Test：0.3082 | Train_AUC Score ：0.5510 Test_AUC Score ：0.5164 |  |
|
| SVM_rbf | Train_准确性：0.7971 Test_准确性：0.7589 | Train_召回率：0.1894 Test_召回率：0.1455 | Train_f1_score：0.3119 Test_f1_score：0.2456 | Train：0.6474 Test：0.3723 | Train_AUC Score ：0.5907 Test_AUC Score ：0.5655 |  |
|
| SVM_sigmoid | Train_准确性：0.7265 Test_准确性：0.7092 | Train_召回率：0.2809 Test_召回率：0.1584 | Train_f1_score：0.3328 Test_f1_score：0.2272 | Train：0.2216 Test：0.1235 | Train_AUC Score ：0.5752 Test_AUC Score ：0.5356 |  |
|
| lg_120 | Train_准确性：0.4355 Test_准确性：0.4590 | Train_召回率：0.6671 Test_召回率：0.7117 | Train_f1_score：0.3647 Test_f1_score：0.4152 | Train：0.0695 Test：0.0907 | Train_AUC Score ：0.5142 Test_AUC Score ：0.5387 |  |
|
| DT | Train_准确性：0.7920 Test_准确性：0.7505 | Train_召回率：0.4245 Test_召回率：0.3169 | Train_f1_score：0.4978 Test_f1_score：0.4067 | Train：0.4126 Test：0.3524 | Train_AUC Score ：0.6672 Test_AUC Score ：0.6138 |  |
|
| xgb_sklearn | Train_准确性：0.8452 Test_准确性：0.7765 | Train_召回率：0.4691 Test_召回率：0.3065 | Train_f1_score：0.5954 Test_f1_score：0.4252 | Train：0.6167 Test：0.3763 | Train_AUC Score ：0.7175 Test_AUC Score ：0.6283 |  |
|
| lgb_sklearn | Train_准确性：1.0000 Test_准确性：0.7680 | Train_召回率：1.0000 Test_召回率：0.3117 | Train_f1_score：1.0000 Test_f1_score：0.4203 | Train：1.0000 Test：0.3761 | Train_AUC Score ：1.0000 Test_AUC Score ：0.6242 |  |
参考文献
ML实操 - 贷款用户逾期情况分析
ML - 贷款用户逾期情况分析
python matplotlib 画图保存图片简单例子
sklearn.metrics中的评估方法介绍（accuracy_score, recall_score, roc_curve, roc_auc_score, confusion_matrix）